The Development of a Real-Time Gauge QC and Gauge-Radar Merging Scheme for England and Wales

Weather Radar measurements allow high resolution, real-time estimates of precipitation at the surface to be made for use in hydrological models and flood warning applications. However, radar estimates are prone to errors due to the vertical profile of reflectivity, beam broadening, variations in the drop size distribution, attenuation and orographic enhancement. Conversely, rain-gauges provide an accurate but spatially sparse measurement of rainfall accumulations. Skilfully combining radar estimates with gauge measurements can produce a high resolution merged product with an accuracy greater than that of the original datasets.

A joint project has recently been undertaken by the Met Office and the Environment Agency to produce a real-time gauge-radar merging scheme for England and Wales. The process combines 15-minute gauge accumulations from around 1000 quality-controlled tipping-bucket rain-gauges with radar data, 3 hours behind the validity time. The geospatial interpolation method used to blend the gauge and radar data is Kriging with External Drift (KED) using a non-parametric variogram [1]. For each pixel in the radar field KED uses the distance to the nearest neighbour gauges, their accumulation values and the positions of these observation points relative to each other (through a variogram) to modify the radar rainfall accumulation value. Previous tests have shown that KED is suitable for a wide range of meteorological conditions, requires a suitably short processing time and has the most robust response to decreases in gauge density when compared to other Kriging schemes such as Ordinary Kriging and Kriging with Radar-based error correction [2].

A key assumption of the KED scheme is that the gauge data provides a “ground truth” value for the rainfall at discrete pixel locations. This requires the gauges to be rigorously quality controlled, in real time, prior to their inclusion in the merging scheme. The gauge data is received in near real-time as 15 minute accumulations with around one-third of the gauges providing supplementary information such as time-of-tip and present weather measurements. Based on these parameters, a number of gauge QC tests have been developed to detect issues such as double tips, frozen precipitation and partially blocked gauges. In addition, spatial checks are applied to compare each gauge with its nearest neighbours. These spatial checks identify cases of spatially uncorrelated readings which may either be due to a faulty gauge or a localised rainfall event. If any QC checks are failed, the gauge reading is left unchanged but a flag and severity level is set to enable a decision to be made on whether or not to use the gauge in the merging process.

During the testing process it was found that the biggest challenge to delivering a merged product within a usable timescale was the variability in polling regimes across the gauge network. The performance of the KED scheme improves with increasing gauge density but a trade-off is required between a timely delivery of the merged product and receiving sufficient gauge data to produce a reliable result. The solution to this dilemma was to run the merging in near-real-time using all gauges available within 3 hours of the validity time and then re-run the merging 24 hours in arrears to generate a “refined” product using all gauges in the network.

The performance of the semi-operational code was assessed during a ten-week period from 8th November 2014 to 17th January 2015. A simple cross-validation method was used to quantify the scheme performance; one-third of the available gauges were removed for use as test gauges with the remaining gauges were used for the merging. The test gauges were then used to calculate various statistics for the 10-week trial period.

Analysis of the results show that the merging scheme improves the quality of the rainfall accumulation values for all meteorological conditions and rainfall intensities. Cross validation analysis of the refined product shows a consistent reduction in the measurement errors with the mean average error being reduced by 0.2mm at medium and high rainfall accumulations (when compared to the original radar data). In addition to this, an improvement of up to 9% in the detection efficiency was seen for the same threshold intensities. These results are based on a reduced merging network (to accommodate the use of cross-validation gauges) and so the true improvement in the skill scores is expected to be even higher.

References:

[1] Velasco-Forero, C.A., Sempere-Torres, D., Cassiraga, E.F. and Gomez-Hernandez, J.J. (2009). A non-parametric automatic blending methodology to estimate rainfall fields from rain gauge and radar data. Advances in Water Resources, 32 pp986-1002

[2] Jewell, S. A. and Gaussiat, N. (2015), An assessment of Kriging-based rain-gauge–radar merging techniques. Q.J.R. Meteorol. Soc.. doi: 10.1002/qj.2522